VIMAGE + NDIS

[Paul B Mahol]

On 9/27/10, Julian Elischer <julian at freebsd.org> wrote:
>   anyone here have NDIS experience?
>
>   On 9/27/10 10:51 AM, Nikos Vassiliadis wrote:
>>  Julian Elischer wrote:
>>>>  #10 0xc0978200 in rt_dispatch (m=0xc764ad00, sa=0x0) at
>>>>  /usr/src/sys/net/rtsock.c:1374
>>>>  1374            if (V_loif)
>>>>  (kgdb) list
>>>>  1369                    }
>>>>  1370                    *(unsigned short *)(tag + 1) = sa->sa_family;
>>>>  1371                    m_tag_prepend(m, tag);
>>>>  1372            }
>>>>  1373    #ifdef VIMAGE
>>>>  1374            if (V_loif)
>>>>  1375                    m->m_pkthdr.rcvif = V_loif;
>>>>  1376            else {
>>>>  1377                    m_freem(m);
>>>>  1378                    return;
>>>>  (kgdb)
>>>>
>>>
>>>  ok so probably there is a code-path to this point that does not first
>>>  set up the current-vnet pointer before doing this.
>>>  what you need to do is to produce a stack-trace so we can see how
>>>  it got here,
>>>  and then we can figure out where on that path we should set the
>>>  pointer.
>>
>>  Here it is:
>>  (kgdb) #0  doadump () at pcpu.h:231
>>  #1  0xc04d48b9 in db_fncall (dummy1=1, dummy2=0, dummy3=-1058443808,
>>      dummy4=0xeef1d720 "") at /usr/src/sys/ddb/db_command.c:548
>>  #2  0xc04d4cb1 in db_command (last_cmdp=0xc0df4bdc, cmd_table=0x0,
>>  dopager=1)
>>      at /usr/src/sys/ddb/db_command.c:445
>>  #3  0xc04d4e0a in db_command_loop () at
>>  /usr/src/sys/ddb/db_command.c:498
>>  #4  0xc04d6d0d in db_trap (type=12, code=0) at
>>  /usr/src/sys/ddb/db_main.c:229
>>  #5  0xc08e17ce in kdb_trap (type=12, code=0, tf=0xeef1d948)
>>      at /usr/src/sys/kern/subr_kdb.c:535
>>  #6  0xc0c0ae7f in trap_fatal (frame=0xeef1d948, eva=24)
>>      at /usr/src/sys/i386/i386/trap.c:929
>>  #7  0xc0c0b140 in trap_pfault (frame=0xeef1d948, usermode=0, eva=24)
>>      at /usr/src/sys/i386/i386/trap.c:851
>>  #8  0xc0c0bad5 in trap (frame=0xeef1d948) at
>>  /usr/src/sys/i386/i386/trap.c:533
>>  #9  0xc0bec9ac in calltrap () at /usr/src/sys/i386/i386/exception.s:166
>>  #10 0xc0978200 in rt_dispatch (m=0xc764ad00, sa=0x0)
>>      at /usr/src/sys/net/rtsock.c:1374
>>  #11 0xc0978864 in rt_ifmsg (ifp=0xc6c3d400) at
>>  /usr/src/sys/net/rtsock.c:1168
>>  #12 0xc76704a1 in ?? ()
>>  #13 0xc6c3d400 in ?? ()
>>  #14 0xeef1daa8 in ?? ()
>>  #15 0xeef1daf4 in ?? ()
>>  #16 0xc769ecb3 in NdisMIndicateStatusComplete (adapter=0xc76b9200)
>>      at /usr/src/sys/modules/ndis/../../compat/ndis/subr_ndis.c:3105
>>  #17 0xc766d8a1 in ?? ()
>>  #18 0xc76b9200 in ?? ()
>>  #19 0xc76c0000 in ?? ()
>>  #20 0xc76f1000 in ?? ()
>>  #21 0xeef1dacc in ?? ()
>>  #22 0xc79d2afd in ndis_bcmwl5_sys_drv_data_start ()
>>     from /boot/modules/bcmwl5_sys.ko
>>  #23 0x006c0000 in ?? ()
>>  #24 0xeef1dbb4 in ?? ()
>>  #25 0xc79dcdac in ndis_bcmwl5_sys_drv_data_start ()
>>     from /boot/modules/bcmwl5_sys.ko
>>  #26 0xc76c0000 in ?? ()
>>  #27 0xc76c0000 in ?? ()
>>  #28 0xc7340800 in ?? ()
>>  #29 0xc086adcc in hardclock_cpu (usermode=7077888)
>>      at /usr/src/sys/kern/kern_clock.c:447
>
>
> whoa!
> hardclock interrupted itself?
>
>>  #30 0xc79d2afd in ndis_bcmwl5_sys_drv_data_start ()
>>     from /boot/modules/bcmwl5_sys.ko
>>  #31 0x006c0000 in ?? ()
>>  #32 0xeef1dbb4 in ?? ()
>>  #33 0xc79dcdac in ndis_bcmwl5_sys_drv_data_start ()
>>     from /boot/modules/bcmwl5_sys.ko
>>  #34 0xc76c0000 in ?? ()
>>  #35 0xc76c0000 in ?? ()
>>  #36 0xc7340800 in ?? ()
>
> hmmm maybe we need to get ndis to put in  a wrapper at this point that
> is called form hardclock and sets the value before going further
>
> I'd have to look at the code to see what happens here..
>
> *wonders who has their fingers in this code*..
>
>>  #37 0xc086adcc in hardclock_cpu (usermode=-949223424)
>>      at /usr/src/sys/kern/kern_clock.c:447
>>  Previous frame inner to this frame (corrupt stack?)
>>  (kgdb)
>>
>>
>>  and the panic, in case it helps:
>>  Fatal trap 12: page fault while in kernel mode
>>  cpuid = 1; apic id = 01
>>  fault virtual address   = 0x18
>>  fault code              = supervisor read, page not present
>>  instruction pointer     = 0x20:0xc0978200
>>  stack pointer           = 0x28:0xeef1d988
>>  frame pointer           = 0x28:0xeef1d9a0
>>  code segment            = base 0x0, limit 0xfffff, type 0x1b
>>                          = DPL 0, pres 1, def32 1, gran 1
>>  processor eflags        = interrupt enabled, resume, IOPL = 0
>>  current process         = 1404 (Windows Workitem 3)
>>  panic: from debugger
>>  cpuid = 1
>>  KDB: stack backtrace:
>>  Physical memory: 2916 MB
>>  Dumping 113 MB: 98 82 66 50 34 18 2
>>
>>  Thanks, Nikos
>>

Please give more background information of this case.
Personally I never tested VIMAGE.

Note that, for unknown reason (must be bug from old days) if_ndis.c calls
rt_ifmsg() on its own via NdisMIndicateStatusComplete(), and I see no
point to do that.